Comparative Analysis of Fingerprint Authentication Using the Convolutional Neural Network (CNN) Method for Two Datasets

Every human’s fingerprint has its own unique biological uniqueness, which is very much needed in identification using biometric technology. The advantages of fingerprint biometric technology verification are immutable, transferable, and very easy to use. The verification process is very easy, fast, and precise. Therefore, this research was conducted with the aim of analyzing comparatively in determining an accurate method of fingerprint authentication. The dataset processed in this study comes from public, private, and augmentation data, totaling 5,496 images. The method used to process the data is a Convolutional Neural Network (CNN) with 6 architectural models, namely Inception V3, Inception ResNet V2, Mobile Net, ResNet 50, ResNet 152, and VGG 16. The dataset is processed for training and validation using 20 epochs and each epoch has 10 iterations. The highest accuracy results obtained are Inception V3 with a value of 95% on public data, Inception ResNet V3 with a value of 95% on augmented public data, Mobile Net with a value of 96% on public data, ResNet 50 with a value of 70% on data from two datasets, ResNet 152 with a value of 65% in the two augmented data sets, and VGG 16 with a value of 98% in the two augmented data sets. Therefore, VGG 16 is recommended in the selected architectural model to authenticate fingerprint data, and this research can be a reference in choosing a model for subsequent research.